Scholarlink

<p>This paper proposes a framework for deploying Tiny Machine Learning (TinyML) on autonomous drones to enhance precision agriculture, focusing on low-power, real-time plant disease detection and efficient path planning in resource- constrained environments. Leveraging TinyML models such as optimized Convolutional Neural Networks (CNNs) on embedded systems like ESP-EYE 32 microcontrollers, achieving high-accuracy disease inspection with minimal computational resources. The system enables drones to autonomously navigate fields and capture images at optimal intervals for plant health analysis. Simulations using Mission Planner demonstrate that path planning and environmental factors, like wind speed and direction, significantly influence drone performance. Rectangular flight paths maximize energy efficiency, covering 6000-square-meter fields within 6–7&#xa0;min and capturing up to 260 images for disease detection. Single-labeled models outperformed bounding box-labeled models, achieving up to 90% detection accuracy. This framework offers a significant step toward fully autonomous, intelligent drones that minimize energy and computational demands, promoting sustainable and efficient agricultural practices.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Drone-Based Precision Agriculture: Performance Optimization of TinyML Models on Edge Devices and Adaptive Path Planning

  • 期刊论文

  • 发表日期：

    2025年1月31日

查看全文
  • Yagna S. H. Annadata,
  • Aiswariya Thazhathethil,
  • Vishnuvaradhan Moganarengam,
  • Tooraj Nikoubin

摘要

This paper proposes a framework for deploying Tiny Machine Learning (TinyML) on autonomous drones to enhance precision agriculture, focusing on low-power, real-time plant disease detection and efficient path planning in resource- constrained environments. Leveraging TinyML models such as optimized Convolutional Neural Networks (CNNs) on embedded systems like ESP-EYE 32 microcontrollers, achieving high-accuracy disease inspection with minimal computational resources. The system enables drones to autonomously navigate fields and capture images at optimal intervals for plant health analysis. Simulations using Mission Planner demonstrate that path planning and environmental factors, like wind speed and direction, significantly influence drone performance. Rectangular flight paths maximize energy efficiency, covering 6000-square-meter fields within 6–7 min and capturing up to 260 images for disease detection. Single-labeled models outperformed bounding box-labeled models, achieving up to 90% detection accuracy. This framework offers a significant step toward fully autonomous, intelligent drones that minimize energy and computational demands, promoting sustainable and efficient agricultural practices.